An Investigation of the Effects of the Numerical Missing Value Imputation Methods for Click-through Rate Estimation Performances

Missing data is a common issue in datasets and can have a significant impact on the performance of machine learning models. Imputation is a popular technique for handling missing data, where the missing values are estimated based on the observed values in the dataset. In this paper, we shallowly investigate the effects of missing data imputation on click-through rate (CTR) estimation and classification for tabular data. We evaluate the performance of deep CTR estimation methods, such as DeepFM, FiBiNET, ONN, and DCN on a short Criteo dataset with missing data. Here, we compare the performance of different imputation methods, such as mean, median, iterative, and kNN based imputation approaches to fill in the missing value. Our results show that missing data imputation can improve the performance of deep CTR estimation methods in terms of AUC, loss, and f1-scores. Our findings may provide insights into the effects of missing data imputation on the accuracy and reliability of machine learning models for tabular data classification and CTR estimation.